wr P ou a i a à 
ei NE a ee EQ DLL 2 2 22 22222 
MULTIDENSITY AND ITS APPLICATION TO LANDSAT IMAGERY 
by 
Dr J-P ROGALA (*) 
IBM Scientific Centre 
36, Avenue Raymond Poincare 
75116 PARIS-FRANCE 
ABSTRACT 
This paper reports the principal conclusions of a study 
of automated regional anlysis of Landsat imagery begun in 1979. 
It defines the basic algorithm "ICAR" (Interpretation Cartographi- 
que Assistee Regionale) which is a computation of density vectors 
in a scanning window. It is spatial analysis of an image after 
each pixel has been classified by traditional methods. This ICAR 
algorithm allows one to map"landscapes" and classify regions 
instead of individual pixels. 
Results of tests using ICAR for soil mapping and geological 
applications are presented and compared with traditional manual 
and visual interpretation. 
The place of such a study in a survey process and the relation 
between the size of the window and the scale of the final map are 
discussed throught eight test areas chosen on different Landsat 
images. 
A satellite image is not a map and has to be further proces- 
sed in order to be used as a map. Sophisticated techniques such as 
multitemporal classification, often used, usually produces "maps" 
of change which are insufficient for regional monitoring. For 
instance, such mapping does not reflect soil or rock types but 
only the probable land cover for each pixel. So the first problem 
is the nature of the classified and identified pixels the other 
problem is that such a classified image is only the first step 
in the mapping process. Because no spatial information has been 
taken into account during the classification, users must themselves 
group the pixels into regions. This procedure raises the question 
of the utility of classification of individual pixels within an 
image. 
This paper presents the main results of two years work. It 
tests the difference between the numerical analysis of Landsat 
imagery generated by simulating a part of the interpreter's beha- 
viour and that produced by the interpreters themselves. The method 
developped here is named Interpretation Cartographique Assistee 
Regionale (ICAR). 
The following experts agreed to collaborate in experimenting 
with these methods both before or after mapping: M. Boutin ( 
Chambre d'Agriculture de l'Indre et Loire), M. Colson (Chambre 
  
  
(*) Now with the Regional Centre for Services in Surveying, Mapping 
and Remote Sensing, P.O.box 18118 Nairobi, KENYA. 
195